Researchers from the University of Texas at Austin are using machine learning to create customized music playlists based on the listener’s mood.

The system, which was developed to challenge current music streaming services, operates on a network of feedback loops. The listener rates a song selected by the system, and that input shapes the selection of the next song, altering the model accordingly.

The system not only selects music based on the listener's input, it also intelligently arranges selections, as a DJ would, so that there is an understandable pattern, and not seemingly arbitrary selections.

The personal DJ has been dubbed DJ-MC, in a nod to a machine learning mechanism the researchers used called Monte Carlo search, wherein results are calculated iteratively, each time applying a different set of random values from the probability functions.

DJ-MC plans 10 songs in advance, but generates tens of thousands of possible sequences. The system then picks the next song based on user feedback of the current song, and the algorithm again picks new selections.

The developers of DJ-MC are Maytal Saar-Tsechansky, professor of information, risk and operations management at the McCombs School of Business; Elad Liebman, a Ph.D. computer science student; and computer science Professor Peter Stone. They believe the program could be used with other types of media, including videos and news stories.

"Learning algorithms don't have taste, they just have data," said Liebman, "You can replace the dataset with anything, as long as people are consuming it in a similar fashion."

Saar-Tsechansky added, "It can work in any case where you're recommending things to humans, experienced in a sequence. It could even be food."

The paper The Right Music at the Right Time: Adaptive Personalized Playlists Based on Sequence Modeling, has been published in the MIS Quarterly.